There’s been some interesting development. I was wondering what made @totkeks’s setup different from everybody else’s and figured it out. The direction in which the pigtails are routed can affect noise level drastically. Even a single cable facing towards the mini PCI-e slot like shown here was enough to raise the noise to -80 dBm. With all cables going in the opposite direction I got -85 dBm.
Ungrounded copper shielding between the stacked PCBs is still effective regardless of pigtail direction. As is populating SFP+ cages. Here’s the breakdown of measurements:
All pigtails facing mini PCI-e slots (noise dBm):
-80 no shield, no sfp
-84 bottom shield
-88 both sfp
-89 bottom shield + both sfp
All pigtails facing away:
-85 no shield, no sfp
-90 bottom shield
-92 both sfp
-92 bottom shield + sfp
Routing pigtails over the problematic area on the other side of the main board seems safe. Still, after putting the R4 back in the stock metal housing with all those measures combined much of the gains were lost — the noise level went up to -83 dBm. That said, I still have to route the pigtails for the 6 GHz band towards the mini PCI-e slots until I get 30 cm cables to play with.
It remains to be seen whether a proper RF shielding will help, but at this point I’m starting to question if the stock metal chassis is worth keeping instead of a 3D-printed one or no chassis at all.
While at it, I took better measurements (some were off before) and updated the model for the shield cans.
shield_can.skp (343.4 KB)
I will explore some more - with my experiments yesterday I’m not 100% sure it is the coax cables running over the PCIe alone. I also have the feeling it could be the antenna itself, depending on the direction it radiates (or received) related to PCIe. I left the device running and tried to move the coax around and also found places where I had -85 dBm, but had cables just over the PCIe port. In one experiment I left all coax “static” but turned around some antennas so it radiates/received in the direction of the PCIe (90° to antenna stem) and suddenly noise felt into the -72 dBm range.
Perhaps it is a combination of all - I have no real knowledge of RF…
Also measuring the metal case (at last on mine): The large main body is grounded (the antenna attachments scratch the metal and therefore ground it) - but the front/back attachments are not grounded when I screw them in. So it can probably not shield everything correctly.
Just to give a better overview, here is how it looks like in my setup. The antennas for 2.4 and 5GHz are towards the top, the 6GHz are towards the bottom.
I tried connecting the front and back panels to ground — didn’t change anything. Removing the back panel improved the noise from -83 to -86 dBm, though.
Yep 1 SFP makes a difference for some reason, wonder what 2 SFPs would do, I took my copper SFP out cause it ran hot waiting for a heatsink and fan, and that’s when I noticed that I couldn’t really get a reliable and somewhat strong signal in the furthest room of my house
In my experience, adding the second SFP helped too. Less so when everything was assembled in the stock metal housing, though. It can be a DAC instead of a transceiver.
Long-time forum reader registered here to share my finding
I’m using bpi-r4 with be14 board in stock alu case; s/w is the latest snapshot to the date.
Playing with bands (actually moving 6GHz band to 89 channel), I’ve got “perfect” -92dBm noise level. Still I don’t see any difference compared -77dBm when using 6GHz on 1st channel. The range is poor – data rate drops quickly, upload suffers.
I’ve replicated your channel selection on the latest OpenWrt build, but it didn’t improve noise level whatsoever. Would you mind sharing your entire /etc/config/network?
I also double checked on the latest snapshot from firmware selector with the same results. Have you, perhaps, purchased your BPI-R4 recently and ended up getting a new revision of the main or BE14 board?
Thank you. What seems to do the trick is setting the 6 GHz band to a narrow channel of 40 or 20 MHz after turning it on in the first place if it wasn’t before. That way I also got the noise level down to -91 from -82 dBm. The 6 GHz radio can then be turned off and the noise level will stay low until the next reboot. I didn’t notice any improvement in throughput or MSS index on the 5 GHz band, though (could be that -91 dBm is not good enough still). And, needless to say, EHT20/40 is useless for 6 GHz. Nevertheless, a very surprising discovery.
I also analyzed the noise levels and came to the conclusion that the values reported in luci for the 2, 5 and 6 GHz radios are in fact all from the 6 GHz radio. They will always have the same value for all three bands.
With a command like “iw dev survey dump” you can see the actual noise levels for the frequencies you are using. They will be different for each band. I suspect that this bug is coming from the fact that the dump command is printing the noise levels for all bands, even if you query only for exmaple the 2 GHz radio device.
That my be the reason you can artifically tune your noise levels in luci by narrowing the 6 GHz band. But it will not have a beneficial effect on the other bands.
